write.c 50.7 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3
/*
 * linux/fs/nfs/write.c
 *
4
 * Write file data over NFS.
L
Linus Torvalds 已提交
5 6 7 8 9 10 11 12 13 14
 *
 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
 */

#include <linux/types.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
#include <linux/file.h>
#include <linux/writeback.h>
15
#include <linux/swap.h>
16
#include <linux/migrate.h>
L
Linus Torvalds 已提交
17 18 19 20 21

#include <linux/sunrpc/clnt.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_mount.h>
#include <linux/nfs_page.h>
22
#include <linux/backing-dev.h>
23
#include <linux/export.h>
24

L
Linus Torvalds 已提交
25 26 27
#include <asm/uaccess.h>

#include "delegation.h"
28
#include "internal.h"
C
Chuck Lever 已提交
29
#include "iostat.h"
30
#include "nfs4_fs.h"
31
#include "fscache.h"
32
#include "pnfs.h"
L
Linus Torvalds 已提交
33

34 35
#include "nfstrace.h"

L
Linus Torvalds 已提交
36 37 38 39 40 41 42 43
#define NFSDBG_FACILITY		NFSDBG_PAGECACHE

#define MIN_POOL_WRITE		(32)
#define MIN_POOL_COMMIT		(4)

/*
 * Local function declarations
 */
44
static void nfs_redirty_request(struct nfs_page *req);
45
static const struct rpc_call_ops nfs_write_common_ops;
46
static const struct rpc_call_ops nfs_commit_ops;
47
static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
48
static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
L
Linus Torvalds 已提交
49

50
static struct kmem_cache *nfs_wdata_cachep;
51
static mempool_t *nfs_wdata_mempool;
52
static struct kmem_cache *nfs_cdata_cachep;
L
Linus Torvalds 已提交
53 54
static mempool_t *nfs_commit_mempool;

55
struct nfs_commit_data *nfs_commitdata_alloc(void)
L
Linus Torvalds 已提交
56
{
57
	struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
58

L
Linus Torvalds 已提交
59 60 61 62 63 64
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
	}
	return p;
}
65
EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
L
Linus Torvalds 已提交
66

67
void nfs_commit_free(struct nfs_commit_data *p)
L
Linus Torvalds 已提交
68 69 70
{
	mempool_free(p, nfs_commit_mempool);
}
71
EXPORT_SYMBOL_GPL(nfs_commit_free);
L
Linus Torvalds 已提交
72

73
struct nfs_write_header *nfs_writehdr_alloc(void)
74
{
75
	struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
76 77

	if (p) {
78 79
		struct nfs_pgio_header *hdr = &p->header;

80
		memset(p, 0, sizeof(*p));
81
		INIT_LIST_HEAD(&hdr->pages);
82 83 84
		INIT_LIST_HEAD(&hdr->rpc_list);
		spin_lock_init(&hdr->lock);
		atomic_set(&hdr->refcnt, 0);
T
Trond Myklebust 已提交
85
		hdr->verf = &p->verf;
86 87 88
	}
	return p;
}
B
Bryan Schumaker 已提交
89
EXPORT_SYMBOL_GPL(nfs_writehdr_alloc);
90

91 92
static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr,
						  unsigned int pagecount)
93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115
{
	struct nfs_write_data *data, *prealloc;

	prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data;
	if (prealloc->header == NULL)
		data = prealloc;
	else
		data = kzalloc(sizeof(*data), GFP_KERNEL);
	if (!data)
		goto out;

	if (nfs_pgarray_set(&data->pages, pagecount)) {
		data->header = hdr;
		atomic_inc(&hdr->refcnt);
	} else {
		if (data != prealloc)
			kfree(data);
		data = NULL;
	}
out:
	return data;
}

116
void nfs_writehdr_free(struct nfs_pgio_header *hdr)
117
{
118 119
	struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header);
	mempool_free(whdr, nfs_wdata_mempool);
120
}
B
Bryan Schumaker 已提交
121
EXPORT_SYMBOL_GPL(nfs_writehdr_free);
122

123
void nfs_writedata_release(struct nfs_write_data *wdata)
L
Linus Torvalds 已提交
124
{
125 126 127
	struct nfs_pgio_header *hdr = wdata->header;
	struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header);

128
	put_nfs_open_context(wdata->args.context);
F
Fred Isaman 已提交
129 130
	if (wdata->pages.pagevec != wdata->pages.page_array)
		kfree(wdata->pages.pagevec);
131
	if (wdata == &write_header->rpc_data) {
132
		wdata->header = NULL;
133 134
		wdata = NULL;
	}
135
	if (atomic_dec_and_test(&hdr->refcnt))
136
		hdr->completion_ops->completion(hdr);
137 138 139 140
	/* Note: we only free the rpc_task after callbacks are done.
	 * See the comment in rpc_free_task() for why
	 */
	kfree(wdata);
L
Linus Torvalds 已提交
141
}
B
Bryan Schumaker 已提交
142
EXPORT_SYMBOL_GPL(nfs_writedata_release);
L
Linus Torvalds 已提交
143

144 145 146 147 148 149 150
static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
{
	ctx->error = error;
	smp_wmb();
	set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
}

151 152
static struct nfs_page *
nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page)
153 154 155
{
	struct nfs_page *req = NULL;

156
	if (PagePrivate(page))
157
		req = (struct nfs_page *)page_private(page);
158 159 160 161 162 163 164 165 166 167
	else if (unlikely(PageSwapCache(page))) {
		struct nfs_page *freq, *t;

		/* Linearly search the commit list for the correct req */
		list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) {
			if (freq->wb_page == page) {
				req = freq;
				break;
			}
		}
168
	}
169 170 171 172

	if (req)
		kref_get(&req->wb_kref);

173 174 175 176 177
	return req;
}

static struct nfs_page *nfs_page_find_request(struct page *page)
{
178
	struct inode *inode = page_file_mapping(page)->host;
179 180
	struct nfs_page *req = NULL;

181
	spin_lock(&inode->i_lock);
182
	req = nfs_page_find_request_locked(NFS_I(inode), page);
183
	spin_unlock(&inode->i_lock);
184 185 186
	return req;
}

L
Linus Torvalds 已提交
187 188 189
/* Adjust the file length if we're writing beyond the end */
static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
{
190
	struct inode *inode = page_file_mapping(page)->host;
191 192
	loff_t end, i_size;
	pgoff_t end_index;
L
Linus Torvalds 已提交
193

194 195 196
	spin_lock(&inode->i_lock);
	i_size = i_size_read(inode);
	end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
197
	if (i_size > 0 && page_file_index(page) < end_index)
198
		goto out;
199
	end = page_file_offset(page) + ((loff_t)offset+count);
L
Linus Torvalds 已提交
200
	if (i_size >= end)
201
		goto out;
L
Linus Torvalds 已提交
202
	i_size_write(inode, end);
203 204 205
	nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
out:
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
206 207
}

208 209 210
/* A writeback failed: mark the page as bad, and invalidate the page cache */
static void nfs_set_pageerror(struct page *page)
{
211
	nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
212 213
}

L
Linus Torvalds 已提交
214 215 216 217 218 219 220 221 222
/* We can set the PG_uptodate flag if we see that a write request
 * covers the full page.
 */
static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
{
	if (PageUptodate(page))
		return;
	if (base != 0)
		return;
223
	if (count != nfs_page_length(page))
L
Linus Torvalds 已提交
224
		return;
225
	SetPageUptodate(page);
L
Linus Torvalds 已提交
226 227 228 229 230
}

static int wb_priority(struct writeback_control *wbc)
{
	if (wbc->for_reclaim)
231
		return FLUSH_HIGHPRI | FLUSH_STABLE;
232
	if (wbc->for_kupdate || wbc->for_background)
233 234
		return FLUSH_LOWPRI | FLUSH_COND_STABLE;
	return FLUSH_COND_STABLE;
L
Linus Torvalds 已提交
235 236
}

237 238 239 240 241 242 243 244 245 246
/*
 * NFS congestion control
 */

int nfs_congestion_kb;

#define NFS_CONGESTION_ON_THRESH 	(nfs_congestion_kb >> (PAGE_SHIFT-10))
#define NFS_CONGESTION_OFF_THRESH	\
	(NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))

247
static void nfs_set_page_writeback(struct page *page)
248
{
249
	struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
250 251
	int ret = test_set_page_writeback(page);

252
	WARN_ON_ONCE(ret != 0);
253

254 255 256 257
	if (atomic_long_inc_return(&nfss->writeback) >
			NFS_CONGESTION_ON_THRESH) {
		set_bdi_congested(&nfss->backing_dev_info,
					BLK_RW_ASYNC);
258 259 260 261 262
	}
}

static void nfs_end_page_writeback(struct page *page)
{
263
	struct inode *inode = page_file_mapping(page)->host;
264 265 266
	struct nfs_server *nfss = NFS_SERVER(inode);

	end_page_writeback(page);
P
Peter Zijlstra 已提交
267
	if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
268
		clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
269 270
}

271
static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
272
{
273
	struct inode *inode = page_file_mapping(page)->host;
274 275 276
	struct nfs_page *req;
	int ret;

277
	spin_lock(&inode->i_lock);
278
	for (;;) {
279
		req = nfs_page_find_request_locked(NFS_I(inode), page);
280 281
		if (req == NULL)
			break;
282
		if (nfs_lock_request(req))
283 284
			break;
		/* Note: If we hold the page lock, as is the case in nfs_writepage,
285
		 *	 then the call to nfs_lock_request() will always
286 287 288
		 *	 succeed provided that someone hasn't already marked the
		 *	 request as dirty (in which case we don't care).
		 */
289
		spin_unlock(&inode->i_lock);
290 291 292 293
		if (!nonblock)
			ret = nfs_wait_on_request(req);
		else
			ret = -EAGAIN;
294 295
		nfs_release_request(req);
		if (ret != 0)
296
			return ERR_PTR(ret);
297
		spin_lock(&inode->i_lock);
298
	}
299
	spin_unlock(&inode->i_lock);
300 301 302 303 304 305 306 307
	return req;
}

/*
 * Find an associated nfs write request, and prepare to flush it out
 * May return an error if the user signalled nfs_wait_on_request().
 */
static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
308
				struct page *page, bool nonblock)
309 310 311 312
{
	struct nfs_page *req;
	int ret = 0;

313
	req = nfs_find_and_lock_request(page, nonblock);
314 315 316 317 318 319
	if (!req)
		goto out;
	ret = PTR_ERR(req);
	if (IS_ERR(req))
		goto out;

320 321
	nfs_set_page_writeback(page);
	WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
322

323
	ret = 0;
324 325
	if (!nfs_pageio_add_request(pgio, req)) {
		nfs_redirty_request(req);
326
		ret = pgio->pg_error;
327
	}
328 329
out:
	return ret;
330 331
}

T
Trond Myklebust 已提交
332
static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
L
Linus Torvalds 已提交
333
{
334
	struct inode *inode = page_file_mapping(page)->host;
335
	int ret;
L
Linus Torvalds 已提交
336

C
Chuck Lever 已提交
337 338 339
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
	nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);

340
	nfs_pageio_cond_complete(pgio, page_file_index(page));
341
	ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
342 343 344 345 346
	if (ret == -EAGAIN) {
		redirty_page_for_writepage(wbc, page);
		ret = 0;
	}
	return ret;
T
Trond Myklebust 已提交
347
}
348

T
Trond Myklebust 已提交
349 350 351 352 353 354 355
/*
 * Write an mmapped page to the server.
 */
static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
{
	struct nfs_pageio_descriptor pgio;
	int err;
356

357 358
	nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc),
				false, &nfs_async_write_completion_ops);
T
Trond Myklebust 已提交
359 360 361 362 363 364 365
	err = nfs_do_writepage(page, wbc, &pgio);
	nfs_pageio_complete(&pgio);
	if (err < 0)
		return err;
	if (pgio.pg_error < 0)
		return pgio.pg_error;
	return 0;
366 367 368 369
}

int nfs_writepage(struct page *page, struct writeback_control *wbc)
{
T
Trond Myklebust 已提交
370
	int ret;
371

T
Trond Myklebust 已提交
372
	ret = nfs_writepage_locked(page, wbc);
L
Linus Torvalds 已提交
373
	unlock_page(page);
T
Trond Myklebust 已提交
374 375 376 377 378 379 380 381 382 383
	return ret;
}

static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
{
	int ret;

	ret = nfs_do_writepage(page, wbc, data);
	unlock_page(page);
	return ret;
L
Linus Torvalds 已提交
384 385 386 387 388
}

int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
	struct inode *inode = mapping->host;
389
	unsigned long *bitlock = &NFS_I(inode)->flags;
390
	struct nfs_pageio_descriptor pgio;
L
Linus Torvalds 已提交
391 392
	int err;

393 394 395 396 397 398
	/* Stop dirtying of new pages while we sync */
	err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
			nfs_wait_bit_killable, TASK_KILLABLE);
	if (err)
		goto out_err;

C
Chuck Lever 已提交
399 400
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);

401 402
	nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
				&nfs_async_write_completion_ops);
T
Trond Myklebust 已提交
403
	err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
404
	nfs_pageio_complete(&pgio);
405 406 407 408 409

	clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
	smp_mb__after_clear_bit();
	wake_up_bit(bitlock, NFS_INO_FLUSHING);

T
Trond Myklebust 已提交
410
	if (err < 0)
411 412 413 414
		goto out_err;
	err = pgio.pg_error;
	if (err < 0)
		goto out_err;
415
	return 0;
416 417
out_err:
	return err;
L
Linus Torvalds 已提交
418 419 420 421 422
}

/*
 * Insert a write request into an inode
 */
F
Fred Isaman 已提交
423
static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
L
Linus Torvalds 已提交
424 425
{
	struct nfs_inode *nfsi = NFS_I(inode);
426 427

	/* Lock the request! */
428
	nfs_lock_request(req);
429 430

	spin_lock(&inode->i_lock);
431
	if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
432
		inode->i_version++;
433 434 435 436 437 438 439 440 441
	/*
	 * Swap-space should not get truncated. Hence no need to plug the race
	 * with invalidate/truncate.
	 */
	if (likely(!PageSwapCache(req->wb_page))) {
		set_bit(PG_MAPPED, &req->wb_flags);
		SetPagePrivate(req->wb_page);
		set_page_private(req->wb_page, (unsigned long)req);
	}
L
Linus Torvalds 已提交
442
	nfsi->npages++;
443
	kref_get(&req->wb_kref);
444
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
445 446 447
}

/*
448
 * Remove a write request from an inode
L
Linus Torvalds 已提交
449 450 451
 */
static void nfs_inode_remove_request(struct nfs_page *req)
{
452
	struct inode *inode = req->wb_context->dentry->d_inode;
L
Linus Torvalds 已提交
453 454
	struct nfs_inode *nfsi = NFS_I(inode);

455
	spin_lock(&inode->i_lock);
456 457 458 459 460
	if (likely(!PageSwapCache(req->wb_page))) {
		set_page_private(req->wb_page, 0);
		ClearPagePrivate(req->wb_page);
		clear_bit(PG_MAPPED, &req->wb_flags);
	}
L
Linus Torvalds 已提交
461
	nfsi->npages--;
462
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
463 464 465
	nfs_release_request(req);
}

466
static void
F
Fred 已提交
467
nfs_mark_request_dirty(struct nfs_page *req)
468 469 470 471
{
	__set_page_dirty_nobuffers(req->wb_page);
}

B
Bryan Schumaker 已提交
472
#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
473 474 475
/**
 * nfs_request_add_commit_list - add request to a commit list
 * @req: pointer to a struct nfs_page
F
Fred Isaman 已提交
476 477
 * @dst: commit list head
 * @cinfo: holds list lock and accounting info
478
 *
F
Fred Isaman 已提交
479
 * This sets the PG_CLEAN bit, updates the cinfo count of
480 481 482
 * number of outstanding requests requiring a commit as well as
 * the MM page stats.
 *
F
Fred Isaman 已提交
483
 * The caller must _not_ hold the cinfo->lock, but must be
484
 * holding the nfs_page lock.
L
Linus Torvalds 已提交
485
 */
486
void
F
Fred Isaman 已提交
487 488
nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
			    struct nfs_commit_info *cinfo)
L
Linus Torvalds 已提交
489
{
490
	set_bit(PG_CLEAN, &(req)->wb_flags);
F
Fred Isaman 已提交
491 492 493 494
	spin_lock(cinfo->lock);
	nfs_list_add_request(req, dst);
	cinfo->mds->ncommit++;
	spin_unlock(cinfo->lock);
495 496
	if (!cinfo->dreq) {
		inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
497
		inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
498 499 500 501
			     BDI_RECLAIMABLE);
		__mark_inode_dirty(req->wb_context->dentry->d_inode,
				   I_DIRTY_DATASYNC);
	}
L
Linus Torvalds 已提交
502
}
503 504 505 506 507
EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);

/**
 * nfs_request_remove_commit_list - Remove request from a commit list
 * @req: pointer to a nfs_page
F
Fred Isaman 已提交
508
 * @cinfo: holds list lock and accounting info
509
 *
F
Fred Isaman 已提交
510
 * This clears the PG_CLEAN bit, and updates the cinfo's count of
511 512 513
 * number of outstanding requests requiring a commit
 * It does not update the MM page stats.
 *
F
Fred Isaman 已提交
514
 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
515 516
 */
void
F
Fred Isaman 已提交
517 518
nfs_request_remove_commit_list(struct nfs_page *req,
			       struct nfs_commit_info *cinfo)
519 520 521 522
{
	if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
		return;
	nfs_list_remove_request(req);
F
Fred Isaman 已提交
523
	cinfo->mds->ncommit--;
524 525 526
}
EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);

F
Fred Isaman 已提交
527 528 529 530 531 532
static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
				      struct inode *inode)
{
	cinfo->lock = &inode->i_lock;
	cinfo->mds = &NFS_I(inode)->commit_info;
	cinfo->ds = pnfs_get_ds_info(inode);
F
Fred Isaman 已提交
533
	cinfo->dreq = NULL;
534
	cinfo->completion_ops = &nfs_commit_completion_ops;
F
Fred Isaman 已提交
535 536 537 538 539 540
}

void nfs_init_cinfo(struct nfs_commit_info *cinfo,
		    struct inode *inode,
		    struct nfs_direct_req *dreq)
{
541 542 543 544
	if (dreq)
		nfs_init_cinfo_from_dreq(cinfo, dreq);
	else
		nfs_init_cinfo_from_inode(cinfo, inode);
F
Fred Isaman 已提交
545 546
}
EXPORT_SYMBOL_GPL(nfs_init_cinfo);
547 548 549 550

/*
 * Add a request to the inode's commit list.
 */
551
void
F
Fred Isaman 已提交
552 553
nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
			struct nfs_commit_info *cinfo)
554
{
F
Fred Isaman 已提交
555
	if (pnfs_mark_request_commit(req, lseg, cinfo))
556
		return;
F
Fred Isaman 已提交
557
	nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
558
}
559

F
Fred Isaman 已提交
560 561 562 563
static void
nfs_clear_page_commit(struct page *page)
{
	dec_zone_page_state(page, NR_UNSTABLE_NFS);
564
	dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE);
F
Fred Isaman 已提交
565 566
}

567
static void
568 569
nfs_clear_request_commit(struct nfs_page *req)
{
570 571
	if (test_bit(PG_CLEAN, &req->wb_flags)) {
		struct inode *inode = req->wb_context->dentry->d_inode;
F
Fred Isaman 已提交
572
		struct nfs_commit_info cinfo;
573

F
Fred Isaman 已提交
574 575 576 577 578
		nfs_init_cinfo_from_inode(&cinfo, inode);
		if (!pnfs_clear_request_commit(req, &cinfo)) {
			spin_lock(cinfo.lock);
			nfs_request_remove_commit_list(req, &cinfo);
			spin_unlock(cinfo.lock);
579
		}
F
Fred Isaman 已提交
580
		nfs_clear_page_commit(req->wb_page);
581 582 583
	}
}

584 585 586
static inline
int nfs_write_need_commit(struct nfs_write_data *data)
{
587
	if (data->verf.committed == NFS_DATA_SYNC)
588 589
		return data->header->lseg == NULL;
	return data->verf.committed != NFS_FILE_SYNC;
590 591 592
}

#else
593 594 595 596 597 598 599 600 601 602 603
static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
				      struct inode *inode)
{
}

void nfs_init_cinfo(struct nfs_commit_info *cinfo,
		    struct inode *inode,
		    struct nfs_direct_req *dreq)
{
}

604
void
F
Fred Isaman 已提交
605 606
nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
			struct nfs_commit_info *cinfo)
607 608 609
{
}

610
static void
611 612 613 614
nfs_clear_request_commit(struct nfs_page *req)
{
}

615 616 617 618 619 620
static inline
int nfs_write_need_commit(struct nfs_write_data *data)
{
	return 0;
}

621 622
#endif

623
static void nfs_write_completion(struct nfs_pgio_header *hdr)
624
{
F
Fred Isaman 已提交
625
	struct nfs_commit_info cinfo;
626 627 628 629
	unsigned long bytes = 0;

	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out;
F
Fred Isaman 已提交
630
	nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
631 632 633 634 635 636 637
	while (!list_empty(&hdr->pages)) {
		struct nfs_page *req = nfs_list_entry(hdr->pages.next);

		bytes += req->wb_bytes;
		nfs_list_remove_request(req);
		if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
		    (hdr->good_bytes < bytes)) {
638
			nfs_set_pageerror(req->wb_page);
639 640 641 642 643 644 645 646
			nfs_context_set_write_error(req->wb_context, hdr->error);
			goto remove_req;
		}
		if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
			nfs_mark_request_dirty(req);
			goto next;
		}
		if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
647
			memcpy(&req->wb_verf, &hdr->verf->verifier, sizeof(req->wb_verf));
F
Fred Isaman 已提交
648
			nfs_mark_request_commit(req, hdr->lseg, &cinfo);
649 650 651 652 653
			goto next;
		}
remove_req:
		nfs_inode_remove_request(req);
next:
654
		nfs_unlock_request(req);
655
		nfs_end_page_writeback(req->wb_page);
656
		nfs_release_request(req);
657 658 659
	}
out:
	hdr->release(hdr);
660
}
L
Linus Torvalds 已提交
661

B
Bryan Schumaker 已提交
662
#if  IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
F
Fred Isaman 已提交
663 664
static unsigned long
nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
665
{
F
Fred Isaman 已提交
666
	return cinfo->mds->ncommit;
F
Fred Isaman 已提交
667 668
}

F
Fred Isaman 已提交
669
/* cinfo->lock held by caller */
670
int
F
Fred Isaman 已提交
671 672
nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
		     struct nfs_commit_info *cinfo, int max)
F
Fred Isaman 已提交
673 674 675 676 677
{
	struct nfs_page *req, *tmp;
	int ret = 0;

	list_for_each_entry_safe(req, tmp, src, wb_list) {
678 679
		if (!nfs_lock_request(req))
			continue;
680
		kref_get(&req->wb_kref);
F
Fred Isaman 已提交
681
		if (cond_resched_lock(cinfo->lock))
682
			list_safe_reset_next(req, tmp, wb_list);
F
Fred Isaman 已提交
683
		nfs_request_remove_commit_list(req, cinfo);
684 685
		nfs_list_add_request(req, dst);
		ret++;
686
		if ((ret == max) && !cinfo->dreq)
687
			break;
F
Fred Isaman 已提交
688 689
	}
	return ret;
690 691
}

L
Linus Torvalds 已提交
692 693 694
/*
 * nfs_scan_commit - Scan an inode for commit requests
 * @inode: NFS inode to scan
F
Fred Isaman 已提交
695 696
 * @dst: mds destination list
 * @cinfo: mds and ds lists of reqs ready to commit
L
Linus Torvalds 已提交
697 698 699 700
 *
 * Moves requests from the inode's 'commit' request list.
 * The requests are *not* checked to ensure that they form a contiguous set.
 */
701
int
F
Fred Isaman 已提交
702 703
nfs_scan_commit(struct inode *inode, struct list_head *dst,
		struct nfs_commit_info *cinfo)
L
Linus Torvalds 已提交
704
{
F
Fred Isaman 已提交
705
	int ret = 0;
706

F
Fred Isaman 已提交
707 708
	spin_lock(cinfo->lock);
	if (cinfo->mds->ncommit > 0) {
709
		const int max = INT_MAX;
F
Fred Isaman 已提交
710

F
Fred Isaman 已提交
711 712 713
		ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
					   cinfo, max);
		ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
F
Fred Isaman 已提交
714
	}
F
Fred Isaman 已提交
715
	spin_unlock(cinfo->lock);
716
	return ret;
L
Linus Torvalds 已提交
717
}
F
Fred Isaman 已提交
718

719
#else
F
Fred Isaman 已提交
720
static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
721 722 723 724
{
	return 0;
}

725 726
int nfs_scan_commit(struct inode *inode, struct list_head *dst,
		    struct nfs_commit_info *cinfo)
727 728 729
{
	return 0;
}
L
Linus Torvalds 已提交
730 731 732
#endif

/*
733 734
 * Search for an existing write request, and attempt to update
 * it to reflect a new dirty region on a given page.
L
Linus Torvalds 已提交
735
 *
736 737
 * If the attempt fails, then the existing request is flushed out
 * to disk.
L
Linus Torvalds 已提交
738
 */
739 740 741 742
static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
		struct page *page,
		unsigned int offset,
		unsigned int bytes)
L
Linus Torvalds 已提交
743
{
744 745 746 747 748 749 750
	struct nfs_page *req;
	unsigned int rqend;
	unsigned int end;
	int error;

	if (!PagePrivate(page))
		return NULL;
L
Linus Torvalds 已提交
751 752

	end = offset + bytes;
753
	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
754 755

	for (;;) {
756
		req = nfs_page_find_request_locked(NFS_I(inode), page);
757 758 759 760 761 762 763 764 765 766
		if (req == NULL)
			goto out_unlock;

		rqend = req->wb_offset + req->wb_bytes;
		/*
		 * Tell the caller to flush out the request if
		 * the offsets are non-contiguous.
		 * Note: nfs_flush_incompatible() will already
		 * have flushed out requests having wrong owners.
		 */
767
		if (offset > rqend
768 769 770
		    || end < req->wb_offset)
			goto out_flushme;

771
		if (nfs_lock_request(req))
L
Linus Torvalds 已提交
772 773
			break;

774
		/* The request is locked, so wait and then retry */
775
		spin_unlock(&inode->i_lock);
776 777 778 779 780
		error = nfs_wait_on_request(req);
		nfs_release_request(req);
		if (error != 0)
			goto out_err;
		spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
781 782 783 784 785 786 787 788 789
	}

	/* Okay, the request matches. Update the region */
	if (offset < req->wb_offset) {
		req->wb_offset = offset;
		req->wb_pgbase = offset;
	}
	if (end > rqend)
		req->wb_bytes = end - req->wb_offset;
790 791 792 793
	else
		req->wb_bytes = rqend - req->wb_offset;
out_unlock:
	spin_unlock(&inode->i_lock);
794 795
	if (req)
		nfs_clear_request_commit(req);
796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
	return req;
out_flushme:
	spin_unlock(&inode->i_lock);
	nfs_release_request(req);
	error = nfs_wb_page(inode, page);
out_err:
	return ERR_PTR(error);
}

/*
 * Try to update an existing write request, or create one if there is none.
 *
 * Note: Should always be called with the Page Lock held to prevent races
 * if we have to add a new request. Also assumes that the caller has
 * already called nfs_flush_incompatible() if necessary.
 */
static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
		struct page *page, unsigned int offset, unsigned int bytes)
{
815
	struct inode *inode = page_file_mapping(page)->host;
816
	struct nfs_page	*req;
L
Linus Torvalds 已提交
817

818 819 820 821 822 823
	req = nfs_try_to_update_request(inode, page, offset, bytes);
	if (req != NULL)
		goto out;
	req = nfs_create_request(ctx, inode, page, offset, bytes);
	if (IS_ERR(req))
		goto out;
F
Fred Isaman 已提交
824
	nfs_inode_add_request(inode, req);
825
out:
T
Trond Myklebust 已提交
826
	return req;
L
Linus Torvalds 已提交
827 828
}

829 830 831 832 833 834 835 836 837 838 839
static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
		unsigned int offset, unsigned int count)
{
	struct nfs_page	*req;

	req = nfs_setup_write_request(ctx, page, offset, count);
	if (IS_ERR(req))
		return PTR_ERR(req);
	/* Update file length */
	nfs_grow_file(page, offset, count);
	nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
840
	nfs_mark_request_dirty(req);
841
	nfs_unlock_and_release_request(req);
842 843 844
	return 0;
}

L
Linus Torvalds 已提交
845 846
int nfs_flush_incompatible(struct file *file, struct page *page)
{
847
	struct nfs_open_context *ctx = nfs_file_open_context(file);
848
	struct nfs_lock_context *l_ctx;
L
Linus Torvalds 已提交
849
	struct nfs_page	*req;
T
Trond Myklebust 已提交
850
	int do_flush, status;
L
Linus Torvalds 已提交
851 852 853 854 855 856 857 858
	/*
	 * Look for a request corresponding to this page. If there
	 * is one, and it belongs to another file, we flush it out
	 * before we try to copy anything into the page. Do this
	 * due to the lack of an ACCESS-type call in NFSv2.
	 * Also do the same if we find a request from an existing
	 * dropped page.
	 */
T
Trond Myklebust 已提交
859 860 861 862
	do {
		req = nfs_page_find_request(page);
		if (req == NULL)
			return 0;
863 864
		l_ctx = req->wb_lock_context;
		do_flush = req->wb_page != page || req->wb_context != ctx;
865
		if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) {
866 867 868
			do_flush |= l_ctx->lockowner.l_owner != current->files
				|| l_ctx->lockowner.l_pid != current->tgid;
		}
L
Linus Torvalds 已提交
869
		nfs_release_request(req);
T
Trond Myklebust 已提交
870 871
		if (!do_flush)
			return 0;
872
		status = nfs_wb_page(page_file_mapping(page)->host, page);
T
Trond Myklebust 已提交
873 874
	} while (status == 0);
	return status;
L
Linus Torvalds 已提交
875 876
}

877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903
/*
 * Avoid buffered writes when a open context credential's key would
 * expire soon.
 *
 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
 *
 * Return 0 and set a credential flag which triggers the inode to flush
 * and performs  NFS_FILE_SYNC writes if the key will expired within
 * RPC_KEY_EXPIRE_TIMEO.
 */
int
nfs_key_timeout_notify(struct file *filp, struct inode *inode)
{
	struct nfs_open_context *ctx = nfs_file_open_context(filp);
	struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;

	return rpcauth_key_timeout_notify(auth, ctx->cred);
}

/*
 * Test if the open context credential key is marked to expire soon.
 */
bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
{
	return rpcauth_cred_key_to_expire(ctx->cred);
}

904 905 906 907 908
/*
 * If the page cache is marked as unsafe or invalid, then we can't rely on
 * the PageUptodate() flag. In this case, we will need to turn off
 * write optimisations that depend on the page contents being correct.
 */
909
static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
910
{
911 912
	struct nfs_inode *nfsi = NFS_I(inode);

913 914
	if (nfs_have_delegated_attributes(inode))
		goto out;
915 916
	if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE))
		return false;
917
	smp_rmb();
918
	if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
919 920 921
		return false;
out:
	return PageUptodate(page) != 0;
922 923
}

924 925 926 927 928
/* If we know the page is up to date, and we're not using byte range locks (or
 * if we have the whole file locked for writing), it may be more efficient to
 * extend the write to cover the entire page in order to avoid fragmentation
 * inefficiencies.
 *
929 930
 * If the file is opened for synchronous writes then we can just skip the rest
 * of the checks.
931 932 933 934 935
 */
static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
{
	if (file->f_flags & O_DSYNC)
		return 0;
936 937
	if (!nfs_write_pageuptodate(page, inode))
		return 0;
938 939
	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
		return 1;
940
	if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 &&
941
			inode->i_flock->fl_end == OFFSET_MAX &&
942
			inode->i_flock->fl_type != F_RDLCK))
943 944 945 946
		return 1;
	return 0;
}

L
Linus Torvalds 已提交
947 948 949 950 951 952 953 954 955
/*
 * Update and possibly write a cached page of an NFS file.
 *
 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
 * things with a page scheduled for an RPC call (e.g. invalidate it).
 */
int nfs_updatepage(struct file *file, struct page *page,
		unsigned int offset, unsigned int count)
{
956
	struct nfs_open_context *ctx = nfs_file_open_context(file);
957
	struct inode	*inode = page_file_mapping(page)->host;
L
Linus Torvalds 已提交
958 959
	int		status = 0;

C
Chuck Lever 已提交
960 961
	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);

962 963
	dprintk("NFS:       nfs_updatepage(%pD2 %d@%lld)\n",
		file, count, (long long)(page_file_offset(page) + offset));
L
Linus Torvalds 已提交
964

965
	if (nfs_can_extend_write(file, page, inode)) {
966
		count = max(count + offset, nfs_page_length(page));
L
Linus Torvalds 已提交
967 968 969
		offset = 0;
	}

970
	status = nfs_writepage_setup(ctx, page, offset, count);
971 972
	if (status < 0)
		nfs_set_pageerror(page);
973 974
	else
		__set_page_dirty_nobuffers(page);
L
Linus Torvalds 已提交
975

976
	dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
L
Linus Torvalds 已提交
977 978 979 980
			status, (long long)i_size_read(inode));
	return status;
}

981
static int flush_task_priority(int how)
L
Linus Torvalds 已提交
982 983 984 985 986 987 988 989 990 991
{
	switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
		case FLUSH_HIGHPRI:
			return RPC_PRIORITY_HIGH;
		case FLUSH_LOWPRI:
			return RPC_PRIORITY_LOW;
	}
	return RPC_PRIORITY_NORMAL;
}

992 993
int nfs_initiate_write(struct rpc_clnt *clnt,
		       struct nfs_write_data *data,
994
		       const struct rpc_call_ops *call_ops,
995
		       int how, int flags)
L
Linus Torvalds 已提交
996
{
997
	struct inode *inode = data->header->inode;
998
	int priority = flush_task_priority(how);
999
	struct rpc_task *task;
1000 1001 1002
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
1003
		.rpc_cred = data->header->cred,
1004
	};
1005
	struct rpc_task_setup task_setup_data = {
1006
		.rpc_client = clnt,
1007
		.task = &data->task,
1008
		.rpc_message = &msg,
1009 1010
		.callback_ops = call_ops,
		.callback_data = data,
1011
		.workqueue = nfsiod_workqueue,
1012
		.flags = RPC_TASK_ASYNC | flags,
1013
		.priority = priority,
1014
	};
1015
	int ret = 0;
L
Linus Torvalds 已提交
1016

1017 1018 1019 1020
	/* Set up the initial task struct.  */
	NFS_PROTO(inode)->write_setup(data, &msg);

	dprintk("NFS: %5u initiated write call "
1021
		"(req %s/%llu, %u bytes @ offset %llu)\n",
1022 1023
		data->task.tk_pid,
		inode->i_sb->s_id,
1024
		(unsigned long long)NFS_FILEID(inode),
1025 1026 1027
		data->args.count,
		(unsigned long long)data->args.offset);

1028 1029 1030
	nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client,
				 &task_setup_data.rpc_client, &msg, data);

1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
	task = rpc_run_task(&task_setup_data);
	if (IS_ERR(task)) {
		ret = PTR_ERR(task);
		goto out;
	}
	if (how & FLUSH_SYNC) {
		ret = rpc_wait_for_completion_task(task);
		if (ret == 0)
			ret = task->tk_status;
	}
	rpc_put_task(task);
out:
	return ret;
}
1045
EXPORT_SYMBOL_GPL(nfs_initiate_write);
1046 1047 1048 1049

/*
 * Set up the argument/result storage required for the RPC call.
 */
1050
static void nfs_write_rpcsetup(struct nfs_write_data *data,
1051
		unsigned int count, unsigned int offset,
F
Fred Isaman 已提交
1052
		int how, struct nfs_commit_info *cinfo)
1053
{
1054
	struct nfs_page *req = data->header->req;
1055

L
Linus Torvalds 已提交
1056 1057 1058
	/* Set up the RPC argument and reply structs
	 * NB: take care not to mess about with data->commit et al. */

1059
	data->args.fh     = NFS_FH(data->header->inode);
L
Linus Torvalds 已提交
1060
	data->args.offset = req_offset(req) + offset;
1061 1062
	/* pnfs_set_layoutcommit needs this */
	data->mds_offset = data->args.offset;
L
Linus Torvalds 已提交
1063
	data->args.pgbase = req->wb_pgbase + offset;
F
Fred Isaman 已提交
1064
	data->args.pages  = data->pages.pagevec;
L
Linus Torvalds 已提交
1065
	data->args.count  = count;
1066
	data->args.context = get_nfs_open_context(req->wb_context);
1067
	data->args.lock_context = req->wb_lock_context;
1068
	data->args.stable  = NFS_UNSTABLE;
1069 1070 1071 1072
	switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
	case 0:
		break;
	case FLUSH_COND_STABLE:
F
Fred Isaman 已提交
1073
		if (nfs_reqs_to_commit(cinfo))
1074 1075 1076
			break;
	default:
		data->args.stable = NFS_FILE_SYNC;
1077
	}
L
Linus Torvalds 已提交
1078 1079 1080 1081

	data->res.fattr   = &data->fattr;
	data->res.count   = count;
	data->res.verf    = &data->verf;
1082
	nfs_fattr_init(&data->fattr);
1083
}
L
Linus Torvalds 已提交
1084

1085 1086 1087 1088
static int nfs_do_write(struct nfs_write_data *data,
		const struct rpc_call_ops *call_ops,
		int how)
{
1089
	struct inode *inode = data->header->inode;
1090

1091
	return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0);
L
Linus Torvalds 已提交
1092 1093
}

1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
static int nfs_do_multiple_writes(struct list_head *head,
		const struct rpc_call_ops *call_ops,
		int how)
{
	struct nfs_write_data *data;
	int ret = 0;

	while (!list_empty(head)) {
		int ret2;

1104
		data = list_first_entry(head, struct nfs_write_data, list);
1105 1106
		list_del_init(&data->list);
		
1107
		ret2 = nfs_do_write(data, call_ops, how);
1108 1109 1110 1111 1112 1113
		 if (ret == 0)
			 ret = ret2;
	}
	return ret;
}

F
Fred 已提交
1114 1115 1116 1117 1118 1119 1120
/* If a nfs_flush_* function fails, it should remove reqs from @head and
 * call this on each, which will prepare them to be retried on next
 * writeback using standard nfs.
 */
static void nfs_redirty_request(struct nfs_page *req)
{
	nfs_mark_request_dirty(req);
1121
	nfs_unlock_request(req);
1122
	nfs_end_page_writeback(req->wb_page);
1123
	nfs_release_request(req);
F
Fred 已提交
1124 1125
}

1126
static void nfs_async_write_error(struct list_head *head)
1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
{
	struct nfs_page	*req;

	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_redirty_request(req);
	}
}

1137 1138 1139 1140 1141
static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
	.error_cleanup = nfs_async_write_error,
	.completion = nfs_write_completion,
};

1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
static void nfs_flush_error(struct nfs_pageio_descriptor *desc,
		struct nfs_pgio_header *hdr)
{
	set_bit(NFS_IOHDR_REDO, &hdr->flags);
	while (!list_empty(&hdr->rpc_list)) {
		struct nfs_write_data *data = list_first_entry(&hdr->rpc_list,
				struct nfs_write_data, list);
		list_del(&data->list);
		nfs_writedata_release(data);
	}
	desc->pg_completion_ops->error_cleanup(&desc->pg_list);
}

L
Linus Torvalds 已提交
1155 1156 1157 1158
/*
 * Generate multiple small requests to write out a single
 * contiguous dirty area on one page.
 */
1159 1160
static int nfs_flush_multi(struct nfs_pageio_descriptor *desc,
			   struct nfs_pgio_header *hdr)
L
Linus Torvalds 已提交
1161
{
1162
	struct nfs_page *req = hdr->req;
L
Linus Torvalds 已提交
1163 1164
	struct page *page = req->wb_page;
	struct nfs_write_data *data;
1165
	size_t wsize = desc->pg_bsize, nbytes;
1166
	unsigned int offset;
L
Linus Torvalds 已提交
1167
	int requests = 0;
F
Fred Isaman 已提交
1168
	struct nfs_commit_info cinfo;
L
Linus Torvalds 已提交
1169

F
Fred Isaman 已提交
1170
	nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
L
Linus Torvalds 已提交
1171

1172
	if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
F
Fred Isaman 已提交
1173
	    (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) ||
1174 1175 1176 1177
	     desc->pg_count > wsize))
		desc->pg_ioflags &= ~FLUSH_COND_STABLE;


1178
	offset = 0;
F
Fred Isaman 已提交
1179
	nbytes = desc->pg_count;
1180 1181 1182
	do {
		size_t len = min(nbytes, wsize);

1183
		data = nfs_writedata_alloc(hdr, 1);
1184 1185 1186 1187
		if (!data) {
			nfs_flush_error(desc, hdr);
			return -ENOMEM;
		}
F
Fred Isaman 已提交
1188
		data->pages.pagevec[0] = page;
F
Fred Isaman 已提交
1189
		nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo);
1190
		list_add(&data->list, &hdr->rpc_list);
L
Linus Torvalds 已提交
1191
		requests++;
1192
		nbytes -= len;
1193
		offset += len;
1194
	} while (nbytes != 0);
1195 1196
	nfs_list_remove_request(req);
	nfs_list_add_request(req, &hdr->pages);
1197
	desc->pg_rpc_callops = &nfs_write_common_ops;
1198
	return 0;
L
Linus Torvalds 已提交
1199 1200 1201 1202 1203 1204 1205 1206 1207 1208
}

/*
 * Create an RPC task for the given write request and kick it.
 * The page must have been locked by the caller.
 *
 * It may happen that the page we're passed is not marked dirty.
 * This is the case if nfs_updatepage detects a conflicting request
 * that has been written but not committed.
 */
1209 1210
static int nfs_flush_one(struct nfs_pageio_descriptor *desc,
			 struct nfs_pgio_header *hdr)
L
Linus Torvalds 已提交
1211 1212 1213 1214
{
	struct nfs_page		*req;
	struct page		**pages;
	struct nfs_write_data	*data;
F
Fred Isaman 已提交
1215
	struct list_head *head = &desc->pg_list;
F
Fred Isaman 已提交
1216
	struct nfs_commit_info cinfo;
L
Linus Torvalds 已提交
1217

1218 1219 1220
	data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base,
							   desc->pg_count));
	if (!data) {
1221 1222
		nfs_flush_error(desc, hdr);
		return -ENOMEM;
1223
	}
1224

F
Fred Isaman 已提交
1225
	nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
F
Fred Isaman 已提交
1226
	pages = data->pages.pagevec;
L
Linus Torvalds 已提交
1227 1228 1229
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
1230
		nfs_list_add_request(req, &hdr->pages);
L
Linus Torvalds 已提交
1231 1232 1233
		*pages++ = req->wb_page;
	}

1234
	if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
F
Fred Isaman 已提交
1235
	    (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
1236 1237
		desc->pg_ioflags &= ~FLUSH_COND_STABLE;

L
Linus Torvalds 已提交
1238
	/* Set up the argument struct */
F
Fred Isaman 已提交
1239
	nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
1240 1241
	list_add(&data->list, &hdr->rpc_list);
	desc->pg_rpc_callops = &nfs_write_common_ops;
1242
	return 0;
L
Linus Torvalds 已提交
1243 1244
}

1245 1246
int nfs_generic_flush(struct nfs_pageio_descriptor *desc,
		      struct nfs_pgio_header *hdr)
1247 1248
{
	if (desc->pg_bsize < PAGE_CACHE_SIZE)
1249 1250
		return nfs_flush_multi(desc, hdr);
	return nfs_flush_one(desc, hdr);
1251
}
B
Bryan Schumaker 已提交
1252
EXPORT_SYMBOL_GPL(nfs_generic_flush);
1253 1254

static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1255
{
1256 1257
	struct nfs_write_header *whdr;
	struct nfs_pgio_header *hdr;
1258 1259
	int ret;

1260 1261
	whdr = nfs_writehdr_alloc();
	if (!whdr) {
1262
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1263 1264 1265 1266 1267 1268
		return -ENOMEM;
	}
	hdr = &whdr->header;
	nfs_pgheader_init(desc, hdr, nfs_writehdr_free);
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_flush(desc, hdr);
1269
	if (ret == 0)
1270 1271 1272 1273
		ret = nfs_do_multiple_writes(&hdr->rpc_list,
					     desc->pg_rpc_callops,
					     desc->pg_ioflags);
	if (atomic_dec_and_test(&hdr->refcnt))
1274
		hdr->completion_ops->completion(hdr);
1275
	return ret;
1276 1277 1278 1279 1280 1281 1282
}

static const struct nfs_pageio_ops nfs_pageio_write_ops = {
	.pg_test = nfs_generic_pg_test,
	.pg_doio = nfs_generic_pg_writepages,
};

1283
void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1284
			       struct inode *inode, int ioflags, bool force_mds,
1285
			       const struct nfs_pgio_completion_ops *compl_ops)
L
Linus Torvalds 已提交
1286
{
1287 1288 1289 1290 1291 1292 1293 1294
	struct nfs_server *server = NFS_SERVER(inode);
	const struct nfs_pageio_ops *pg_ops = &nfs_pageio_write_ops;

#ifdef CONFIG_NFS_V4_1
	if (server->pnfs_curr_ld && !force_mds)
		pg_ops = server->pnfs_curr_ld->pg_write_ops;
#endif
	nfs_pageio_init(pgio, inode, pg_ops, compl_ops, server->wsize, ioflags);
1295
}
B
Bryan Schumaker 已提交
1296
EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
L
Linus Torvalds 已提交
1297

1298 1299 1300 1301 1302
void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
{
	pgio->pg_ops = &nfs_pageio_write_ops;
	pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
}
1303
EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1304

L
Linus Torvalds 已提交
1305

1306 1307 1308
void nfs_write_prepare(struct rpc_task *task, void *calldata)
{
	struct nfs_write_data *data = calldata;
1309 1310 1311 1312
	int err;
	err = NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data);
	if (err)
		rpc_exit(task, err);
1313 1314
}

1315 1316 1317 1318 1319 1320 1321
void nfs_commit_prepare(struct rpc_task *task, void *calldata)
{
	struct nfs_commit_data *data = calldata;

	NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
}

L
Linus Torvalds 已提交
1322 1323 1324 1325 1326 1327 1328
/*
 * Handle a write reply that flushes a whole page.
 *
 * FIXME: There is an inherent race with invalidate_inode_pages and
 *	  writebacks since the page->count is kept > 1 for as long
 *	  as the page has a write request pending.
 */
1329
static void nfs_writeback_done_common(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1330
{
1331
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1332

1333 1334 1335
	nfs_writeback_done(task, data);
}

1336
static void nfs_writeback_release_common(void *calldata)
1337 1338
{
	struct nfs_write_data	*data = calldata;
1339
	struct nfs_pgio_header *hdr = data->header;
1340
	int status = data->task.tk_status;
1341

1342 1343 1344 1345 1346
	if ((status >= 0) && nfs_write_need_commit(data)) {
		spin_lock(&hdr->lock);
		if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags))
			; /* Do nothing */
		else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags))
T
Trond Myklebust 已提交
1347 1348
			memcpy(hdr->verf, &data->verf, sizeof(*hdr->verf));
		else if (memcmp(hdr->verf, &data->verf, sizeof(*hdr->verf)))
1349 1350
			set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags);
		spin_unlock(&hdr->lock);
L
Linus Torvalds 已提交
1351
	}
1352
	nfs_writedata_release(data);
L
Linus Torvalds 已提交
1353 1354
}

1355
static const struct rpc_call_ops nfs_write_common_ops = {
1356
	.rpc_call_prepare = nfs_write_prepare,
1357 1358
	.rpc_call_done = nfs_writeback_done_common,
	.rpc_release = nfs_writeback_release_common,
1359 1360
};

1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384
/*
 * Special version of should_remove_suid() that ignores capabilities.
 */
static int nfs_should_remove_suid(const struct inode *inode)
{
	umode_t mode = inode->i_mode;
	int kill = 0;

	/* suid always must be killed */
	if (unlikely(mode & S_ISUID))
		kill = ATTR_KILL_SUID;

	/*
	 * sgid without any exec bits is just a mandatory locking mark; leave
	 * it alone.  If some exec bits are set, it's a real sgid; kill it.
	 */
	if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
		kill |= ATTR_KILL_SGID;

	if (unlikely(kill && S_ISREG(mode)))
		return kill;

	return 0;
}
1385

L
Linus Torvalds 已提交
1386 1387 1388
/*
 * This function is called when the WRITE call is complete.
 */
1389
void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
L
Linus Torvalds 已提交
1390
{
1391
	struct nfs_pgio_args	*argp = &data->args;
L
Linus Torvalds 已提交
1392
	struct nfs_writeres	*resp = &data->res;
1393
	struct inode		*inode = data->header->inode;
1394
	int status;
L
Linus Torvalds 已提交
1395

C
Chuck Lever 已提交
1396
	dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
L
Linus Torvalds 已提交
1397 1398
		task->tk_pid, task->tk_status);

1399 1400 1401 1402 1403 1404 1405
	/*
	 * ->write_done will attempt to use post-op attributes to detect
	 * conflicting writes by other clients.  A strict interpretation
	 * of close-to-open would allow us to continue caching even if
	 * another writer had changed the file, but some applications
	 * depend on tighter cache coherency when writing.
	 */
1406
	status = NFS_PROTO(inode)->write_done(task, data);
1407
	if (status != 0)
1408
		return;
1409
	nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
C
Chuck Lever 已提交
1410

B
Bryan Schumaker 已提交
1411
#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
L
Linus Torvalds 已提交
1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422
	if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
		/* We tried a write call, but the server did not
		 * commit data to stable storage even though we
		 * requested it.
		 * Note: There is a known bug in Tru64 < 5.0 in which
		 *	 the server reports NFS_DATA_SYNC, but performs
		 *	 NFS_FILE_SYNC. We therefore implement this checking
		 *	 as a dprintk() in order to avoid filling syslog.
		 */
		static unsigned long    complain;

1423
		/* Note this will print the MDS for a DS write */
L
Linus Torvalds 已提交
1424
		if (time_before(complain, jiffies)) {
1425
			dprintk("NFS:       faulty NFS server %s:"
L
Linus Torvalds 已提交
1426
				" (committed = %d) != (stable = %d)\n",
1427
				NFS_SERVER(inode)->nfs_client->cl_hostname,
L
Linus Torvalds 已提交
1428 1429 1430 1431 1432
				resp->verf->committed, argp->stable);
			complain = jiffies + 300 * HZ;
		}
	}
#endif
1433
	if (task->tk_status < 0) {
1434
		nfs_set_pgio_error(data->header, task->tk_status, argp->offset);
1435 1436 1437 1438 1439 1440 1441 1442
		return;
	}

	/* Deal with the suid/sgid bit corner case */
	if (nfs_should_remove_suid(inode))
		nfs_mark_for_revalidate(inode);

	if (resp->count < argp->count) {
L
Linus Torvalds 已提交
1443 1444
		static unsigned long    complain;

1445
		/* This a short write! */
1446
		nfs_inc_stats(inode, NFSIOS_SHORTWRITE);
C
Chuck Lever 已提交
1447

L
Linus Torvalds 已提交
1448
		/* Has the server at least made some progress? */
1449 1450 1451 1452 1453 1454
		if (resp->count == 0) {
			if (time_before(complain, jiffies)) {
				printk(KERN_WARNING
				       "NFS: Server wrote zero bytes, expected %u.\n",
				       argp->count);
				complain = jiffies + 300 * HZ;
L
Linus Torvalds 已提交
1455
			}
1456 1457
			nfs_set_pgio_error(data->header, -EIO, argp->offset);
			task->tk_status = -EIO;
1458
			return;
L
Linus Torvalds 已提交
1459
		}
1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471
		/* Was this an NFSv2 write or an NFSv3 stable write? */
		if (resp->verf->committed != NFS_UNSTABLE) {
			/* Resend from where the server left off */
			data->mds_offset += resp->count;
			argp->offset += resp->count;
			argp->pgbase += resp->count;
			argp->count -= resp->count;
		} else {
			/* Resend as a stable write in order to avoid
			 * headaches in the case of a server crash.
			 */
			argp->stable = NFS_FILE_SYNC;
L
Linus Torvalds 已提交
1472
		}
1473
		rpc_restart_call_prepare(task);
L
Linus Torvalds 已提交
1474 1475 1476 1477
	}
}


B
Bryan Schumaker 已提交
1478
#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1479 1480
static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
{
1481 1482
	int ret;

1483 1484
	if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
		return 1;
1485 1486 1487 1488 1489 1490 1491
	if (!may_wait)
		return 0;
	ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
				NFS_INO_COMMIT,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
	return (ret < 0) ? ret : 1;
1492 1493
}

1494
static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
1495 1496 1497 1498 1499 1500
{
	clear_bit(NFS_INO_COMMIT, &nfsi->flags);
	smp_mb__after_clear_bit();
	wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
}

1501
void nfs_commitdata_release(struct nfs_commit_data *data)
L
Linus Torvalds 已提交
1502
{
1503 1504
	put_nfs_open_context(data->context);
	nfs_commit_free(data);
L
Linus Torvalds 已提交
1505
}
1506
EXPORT_SYMBOL_GPL(nfs_commitdata_release);
L
Linus Torvalds 已提交
1507

1508
int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1509
			const struct rpc_call_ops *call_ops,
1510
			int how, int flags)
L
Linus Torvalds 已提交
1511
{
1512
	struct rpc_task *task;
1513
	int priority = flush_task_priority(how);
1514 1515 1516
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
1517
		.rpc_cred = data->cred,
1518
	};
1519
	struct rpc_task_setup task_setup_data = {
1520
		.task = &data->task,
1521
		.rpc_client = clnt,
1522
		.rpc_message = &msg,
1523
		.callback_ops = call_ops,
1524
		.callback_data = data,
1525
		.workqueue = nfsiod_workqueue,
1526
		.flags = RPC_TASK_ASYNC | flags,
1527
		.priority = priority,
1528
	};
1529 1530 1531 1532 1533
	/* Set up the initial task struct.  */
	NFS_PROTO(data->inode)->commit_setup(data, &msg);

	dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);

1534 1535 1536
	nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
		NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);

1537 1538 1539 1540 1541 1542 1543 1544
	task = rpc_run_task(&task_setup_data);
	if (IS_ERR(task))
		return PTR_ERR(task);
	if (how & FLUSH_SYNC)
		rpc_wait_for_completion_task(task);
	rpc_put_task(task);
	return 0;
}
1545
EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1546 1547 1548 1549

/*
 * Set up the argument/result storage required for the RPC call.
 */
1550
void nfs_init_commit(struct nfs_commit_data *data,
1551 1552 1553
		     struct list_head *head,
		     struct pnfs_layout_segment *lseg,
		     struct nfs_commit_info *cinfo)
1554 1555
{
	struct nfs_page *first = nfs_list_entry(head->next);
1556
	struct inode *inode = first->wb_context->dentry->d_inode;
L
Linus Torvalds 已提交
1557 1558 1559 1560 1561 1562 1563

	/* Set up the RPC argument and reply structs
	 * NB: take care not to mess about with data->commit et al. */

	list_splice_init(head, &data->pages);

	data->inode	  = inode;
1564
	data->cred	  = first->wb_context->cred;
1565
	data->lseg	  = lseg; /* reference transferred */
1566
	data->mds_ops     = &nfs_commit_ops;
1567
	data->completion_ops = cinfo->completion_ops;
F
Fred Isaman 已提交
1568
	data->dreq	  = cinfo->dreq;
L
Linus Torvalds 已提交
1569 1570

	data->args.fh     = NFS_FH(data->inode);
1571 1572 1573
	/* Note: we always request a commit of the entire inode */
	data->args.offset = 0;
	data->args.count  = 0;
1574
	data->context     = get_nfs_open_context(first->wb_context);
L
Linus Torvalds 已提交
1575 1576
	data->res.fattr   = &data->fattr;
	data->res.verf    = &data->verf;
1577
	nfs_fattr_init(&data->fattr);
L
Linus Torvalds 已提交
1578
}
1579
EXPORT_SYMBOL_GPL(nfs_init_commit);
L
Linus Torvalds 已提交
1580

1581
void nfs_retry_commit(struct list_head *page_list,
F
Fred Isaman 已提交
1582 1583
		      struct pnfs_layout_segment *lseg,
		      struct nfs_commit_info *cinfo)
1584 1585 1586 1587 1588 1589
{
	struct nfs_page *req;

	while (!list_empty(page_list)) {
		req = nfs_list_entry(page_list->next);
		nfs_list_remove_request(req);
F
Fred Isaman 已提交
1590
		nfs_mark_request_commit(req, lseg, cinfo);
1591 1592
		if (!cinfo->dreq) {
			dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1593
			dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
1594 1595
				     BDI_RECLAIMABLE);
		}
1596
		nfs_unlock_and_release_request(req);
1597 1598
	}
}
1599
EXPORT_SYMBOL_GPL(nfs_retry_commit);
1600

L
Linus Torvalds 已提交
1601 1602 1603 1604
/*
 * Commit dirty pages
 */
static int
F
Fred Isaman 已提交
1605 1606
nfs_commit_list(struct inode *inode, struct list_head *head, int how,
		struct nfs_commit_info *cinfo)
L
Linus Torvalds 已提交
1607
{
1608
	struct nfs_commit_data	*data;
L
Linus Torvalds 已提交
1609

1610
	data = nfs_commitdata_alloc();
L
Linus Torvalds 已提交
1611 1612 1613 1614 1615

	if (!data)
		goto out_bad;

	/* Set up the argument struct */
1616 1617
	nfs_init_commit(data, head, NULL, cinfo);
	atomic_inc(&cinfo->mds->rpcs_out);
1618 1619
	return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops,
				   how, 0);
L
Linus Torvalds 已提交
1620
 out_bad:
F
Fred Isaman 已提交
1621
	nfs_retry_commit(head, NULL, cinfo);
1622
	cinfo->completion_ops->error_cleanup(NFS_I(inode));
L
Linus Torvalds 已提交
1623 1624 1625 1626 1627 1628
	return -ENOMEM;
}

/*
 * COMMIT call returned
 */
1629
static void nfs_commit_done(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1630
{
1631
	struct nfs_commit_data	*data = calldata;
L
Linus Torvalds 已提交
1632

C
Chuck Lever 已提交
1633
        dprintk("NFS: %5u nfs_commit_done (status %d)\n",
L
Linus Torvalds 已提交
1634 1635
                                task->tk_pid, task->tk_status);

1636
	/* Call the NFS version-specific code */
1637
	NFS_PROTO(data->inode)->commit_done(task, data);
1638 1639
}

1640
static void nfs_commit_release_pages(struct nfs_commit_data *data)
1641
{
1642
	struct nfs_page	*req;
1643
	int status = data->task.tk_status;
1644
	struct nfs_commit_info cinfo;
1645

L
Linus Torvalds 已提交
1646 1647 1648
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);
F
Fred Isaman 已提交
1649
		nfs_clear_page_commit(req->wb_page);
L
Linus Torvalds 已提交
1650

1651
		dprintk("NFS:       commit (%s/%llu %d@%lld)",
1652
			req->wb_context->dentry->d_sb->s_id,
1653
			(unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
L
Linus Torvalds 已提交
1654 1655
			req->wb_bytes,
			(long long)req_offset(req));
1656 1657
		if (status < 0) {
			nfs_context_set_write_error(req->wb_context, status);
L
Linus Torvalds 已提交
1658
			nfs_inode_remove_request(req);
1659
			dprintk(", error = %d\n", status);
L
Linus Torvalds 已提交
1660 1661 1662 1663 1664
			goto next;
		}

		/* Okay, COMMIT succeeded, apparently. Check the verifier
		 * returned by the server against all stored verfs. */
1665
		if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
L
Linus Torvalds 已提交
1666 1667 1668 1669 1670 1671 1672
			/* We have a match */
			nfs_inode_remove_request(req);
			dprintk(" OK\n");
			goto next;
		}
		/* We have a mismatch. Write the page again */
		dprintk(" mismatch\n");
F
Fred 已提交
1673
		nfs_mark_request_dirty(req);
1674
		set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
L
Linus Torvalds 已提交
1675
	next:
1676
		nfs_unlock_and_release_request(req);
L
Linus Torvalds 已提交
1677
	}
1678 1679 1680
	nfs_init_cinfo(&cinfo, data->inode, data->dreq);
	if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
		nfs_commit_clear_lock(NFS_I(data->inode));
1681 1682 1683 1684
}

static void nfs_commit_release(void *calldata)
{
1685
	struct nfs_commit_data *data = calldata;
1686

1687
	data->completion_ops->completion(data);
1688
	nfs_commitdata_release(calldata);
L
Linus Torvalds 已提交
1689
}
1690 1691

static const struct rpc_call_ops nfs_commit_ops = {
1692
	.rpc_call_prepare = nfs_commit_prepare,
1693 1694 1695
	.rpc_call_done = nfs_commit_done,
	.rpc_release = nfs_commit_release,
};
L
Linus Torvalds 已提交
1696

1697 1698 1699 1700 1701
static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
	.completion = nfs_commit_release_pages,
	.error_cleanup = nfs_commit_clear_lock,
};

1702 1703
int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
			    int how, struct nfs_commit_info *cinfo)
1704 1705 1706
{
	int status;

F
Fred Isaman 已提交
1707
	status = pnfs_commit_list(inode, head, how, cinfo);
1708
	if (status == PNFS_NOT_ATTEMPTED)
F
Fred Isaman 已提交
1709
		status = nfs_commit_list(inode, head, how, cinfo);
1710 1711 1712
	return status;
}

1713
int nfs_commit_inode(struct inode *inode, int how)
L
Linus Torvalds 已提交
1714 1715
{
	LIST_HEAD(head);
F
Fred Isaman 已提交
1716
	struct nfs_commit_info cinfo;
1717
	int may_wait = how & FLUSH_SYNC;
1718
	int res;
L
Linus Torvalds 已提交
1719

1720 1721
	res = nfs_commit_set_lock(NFS_I(inode), may_wait);
	if (res <= 0)
1722
		goto out_mark_dirty;
F
Fred Isaman 已提交
1723 1724
	nfs_init_cinfo_from_inode(&cinfo, inode);
	res = nfs_scan_commit(inode, &head, &cinfo);
L
Linus Torvalds 已提交
1725
	if (res) {
1726 1727
		int error;

F
Fred Isaman 已提交
1728
		error = nfs_generic_commit_list(inode, &head, how, &cinfo);
1729 1730
		if (error < 0)
			return error;
1731
		if (!may_wait)
1732
			goto out_mark_dirty;
1733 1734 1735 1736 1737 1738
		error = wait_on_bit(&NFS_I(inode)->flags,
				NFS_INO_COMMIT,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
		if (error < 0)
			return error;
1739 1740
	} else
		nfs_commit_clear_lock(NFS_I(inode));
1741 1742 1743 1744 1745 1746 1747 1748
	return res;
	/* Note: If we exit without ensuring that the commit is complete,
	 * we must mark the inode as dirty. Otherwise, future calls to
	 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
	 * that the data is on the disk.
	 */
out_mark_dirty:
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
L
Linus Torvalds 已提交
1749 1750
	return res;
}
1751 1752 1753

static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
1754 1755 1756
	struct nfs_inode *nfsi = NFS_I(inode);
	int flags = FLUSH_SYNC;
	int ret = 0;
1757

1758
	/* no commits means nothing needs to be done */
F
Fred Isaman 已提交
1759
	if (!nfsi->commit_info.ncommit)
1760 1761
		return ret;

1762 1763 1764 1765
	if (wbc->sync_mode == WB_SYNC_NONE) {
		/* Don't commit yet if this is a non-blocking flush and there
		 * are a lot of outstanding writes for this mapping.
		 */
F
Fred Isaman 已提交
1766
		if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1))
1767
			goto out_mark_dirty;
1768

1769
		/* don't wait for the COMMIT response */
1770
		flags = 0;
1771 1772
	}

1773 1774 1775 1776 1777 1778 1779 1780
	ret = nfs_commit_inode(inode, flags);
	if (ret >= 0) {
		if (wbc->sync_mode == WB_SYNC_NONE) {
			if (ret < wbc->nr_to_write)
				wbc->nr_to_write -= ret;
			else
				wbc->nr_to_write = 0;
		}
1781
		return 0;
1782 1783
	}
out_mark_dirty:
1784 1785 1786
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
	return ret;
}
1787
#else
1788 1789 1790 1791
static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
	return 0;
}
L
Linus Torvalds 已提交
1792 1793
#endif

1794 1795
int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
{
1796 1797
	return nfs_commit_unstable_pages(inode, wbc);
}
B
Bryan Schumaker 已提交
1798
EXPORT_SYMBOL_GPL(nfs_write_inode);
1799

1800 1801 1802 1803
/*
 * flush the inode to disk.
 */
int nfs_wb_all(struct inode *inode)
T
Trond Myklebust 已提交
1804 1805
{
	struct writeback_control wbc = {
1806
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1807
		.nr_to_write = LONG_MAX,
1808 1809
		.range_start = 0,
		.range_end = LLONG_MAX,
T
Trond Myklebust 已提交
1810
	};
1811 1812 1813 1814 1815
	int ret;

	trace_nfs_writeback_inode_enter(inode);

	ret = sync_inode(inode, &wbc);
T
Trond Myklebust 已提交
1816

1817 1818
	trace_nfs_writeback_inode_exit(inode, ret);
	return ret;
1819
}
B
Bryan Schumaker 已提交
1820
EXPORT_SYMBOL_GPL(nfs_wb_all);
1821

1822 1823 1824 1825 1826 1827
int nfs_wb_page_cancel(struct inode *inode, struct page *page)
{
	struct nfs_page *req;
	int ret = 0;

	for (;;) {
1828
		wait_on_page_writeback(page);
1829 1830 1831
		req = nfs_page_find_request(page);
		if (req == NULL)
			break;
1832
		if (nfs_lock_request(req)) {
1833
			nfs_clear_request_commit(req);
1834 1835 1836 1837 1838 1839
			nfs_inode_remove_request(req);
			/*
			 * In case nfs_inode_remove_request has marked the
			 * page as being dirty
			 */
			cancel_dirty_page(page, PAGE_CACHE_SIZE);
1840
			nfs_unlock_and_release_request(req);
1841 1842 1843
			break;
		}
		ret = nfs_wait_on_request(req);
1844
		nfs_release_request(req);
1845
		if (ret < 0)
1846
			break;
1847 1848 1849 1850
	}
	return ret;
}

T
Trond Myklebust 已提交
1851 1852 1853 1854
/*
 * Write back all requests on one page - we do this before reading it.
 */
int nfs_wb_page(struct inode *inode, struct page *page)
1855
{
1856
	loff_t range_start = page_file_offset(page);
1857
	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1858 1859
	struct writeback_control wbc = {
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1860
		.nr_to_write = 0,
1861 1862 1863 1864
		.range_start = range_start,
		.range_end = range_end,
	};
	int ret;
1865

1866 1867
	trace_nfs_writeback_page_enter(inode);

1868
	for (;;) {
1869
		wait_on_page_writeback(page);
1870 1871 1872 1873
		if (clear_page_dirty_for_io(page)) {
			ret = nfs_writepage_locked(page, &wbc);
			if (ret < 0)
				goto out_error;
1874
			continue;
T
Trond Myklebust 已提交
1875
		}
1876
		ret = 0;
1877 1878 1879
		if (!PagePrivate(page))
			break;
		ret = nfs_commit_inode(inode, FLUSH_SYNC);
1880
		if (ret < 0)
1881
			goto out_error;
T
Trond Myklebust 已提交
1882
	}
1883
out_error:
1884
	trace_nfs_writeback_page_exit(inode, ret);
1885
	return ret;
1886 1887
}

1888 1889
#ifdef CONFIG_MIGRATION
int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1890
		struct page *page, enum migrate_mode mode)
1891
{
1892 1893 1894 1895 1896 1897 1898 1899 1900 1901
	/*
	 * If PagePrivate is set, then the page is currently associated with
	 * an in-progress read or write request. Don't try to migrate it.
	 *
	 * FIXME: we could do this in principle, but we'll need a way to ensure
	 *        that we can safely release the inode reference while holding
	 *        the page lock.
	 */
	if (PagePrivate(page))
		return -EBUSY;
1902

1903 1904
	if (!nfs_fscache_release_page(page, GFP_KERNEL))
		return -EBUSY;
1905

1906
	return migrate_page(mapping, newpage, page, mode);
1907 1908 1909
}
#endif

D
David Howells 已提交
1910
int __init nfs_init_writepagecache(void)
L
Linus Torvalds 已提交
1911 1912
{
	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1913
					     sizeof(struct nfs_write_header),
L
Linus Torvalds 已提交
1914
					     0, SLAB_HWCACHE_ALIGN,
1915
					     NULL);
L
Linus Torvalds 已提交
1916 1917 1918
	if (nfs_wdata_cachep == NULL)
		return -ENOMEM;

1919 1920
	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
						     nfs_wdata_cachep);
L
Linus Torvalds 已提交
1921
	if (nfs_wdata_mempool == NULL)
1922
		goto out_destroy_write_cache;
L
Linus Torvalds 已提交
1923

1924 1925 1926 1927 1928
	nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
					     sizeof(struct nfs_commit_data),
					     0, SLAB_HWCACHE_ALIGN,
					     NULL);
	if (nfs_cdata_cachep == NULL)
1929
		goto out_destroy_write_mempool;
1930

1931
	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1932
						      nfs_cdata_cachep);
L
Linus Torvalds 已提交
1933
	if (nfs_commit_mempool == NULL)
1934
		goto out_destroy_commit_cache;
L
Linus Torvalds 已提交
1935

1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955
	/*
	 * NFS congestion size, scale with available memory.
	 *
	 *  64MB:    8192k
	 * 128MB:   11585k
	 * 256MB:   16384k
	 * 512MB:   23170k
	 *   1GB:   32768k
	 *   2GB:   46340k
	 *   4GB:   65536k
	 *   8GB:   92681k
	 *  16GB:  131072k
	 *
	 * This allows larger machines to have larger/more transfers.
	 * Limit the default to 256M
	 */
	nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
	if (nfs_congestion_kb > 256*1024)
		nfs_congestion_kb = 256*1024;

L
Linus Torvalds 已提交
1956
	return 0;
1957 1958 1959 1960 1961 1962 1963 1964

out_destroy_commit_cache:
	kmem_cache_destroy(nfs_cdata_cachep);
out_destroy_write_mempool:
	mempool_destroy(nfs_wdata_mempool);
out_destroy_write_cache:
	kmem_cache_destroy(nfs_wdata_cachep);
	return -ENOMEM;
L
Linus Torvalds 已提交
1965 1966
}

1967
void nfs_destroy_writepagecache(void)
L
Linus Torvalds 已提交
1968 1969
{
	mempool_destroy(nfs_commit_mempool);
1970
	kmem_cache_destroy(nfs_cdata_cachep);
L
Linus Torvalds 已提交
1971
	mempool_destroy(nfs_wdata_mempool);
1972
	kmem_cache_destroy(nfs_wdata_cachep);
L
Linus Torvalds 已提交
1973 1974
}